Predict Soybean Yield in Argentina Using MODIS Data Deep Learning Methods

Keywords: Deep Learning, Soybean Yield, Argentina, land cover
Abstract Type: Paper Abstract

Abstract

Crop production is one of the most critical indicators of how agricultural activities are performed each season; it profoundly affects human society. Although making a reliable, timely yield prediction is essential for crop mapping, crop market planning, and harvest management, it remains challenging. Soybean has been one of the essential agricultural products as a source of protein in human history. While Argentina has a long history of soybean production, and its soybean plantation experienced a dramatic increase around the entire country due to the innovation of agricultural and international market demand. We aim to use both a simple linear regression and deep learning model to predict the soybean yield; we hope to use the remote sensing images captured during the growing season at the county level as inputs to get satisfactory results. With the newest land classification products as land cover filters with deep learning methods, the predicted yield result should be higher than traditional regression results. The planned dataset in this project includes a MODIS 16-day vegetation index product, the historical Argentina county-level yield data and the South America soybean land classification product